lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/llite/llite_lib.c
  40  *
  41  * Lustre Light Super operations
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LLITE
  45
  46 #include <linux/module.h>
  47 #include <linux/types.h>
  48 #include <linux/version.h>
  49 #include <linux/mm.h>
  50
  51 #include <lustre_lite.h>
  52 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 cfs_mem_cache_t *ll_file_data_slab;
  63
  64 CFS_LIST_HEAD(ll_super_blocks);
  65 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  66
  67 #ifndef MS_HAS_NEW_AOPS
  68 extern struct address_space_operations ll_aops;
  69 extern struct address_space_operations ll_dir_aops;
  70 #else
  71 extern struct address_space_operations_ext ll_aops;
  72 extern struct address_space_operations_ext ll_dir_aops;
  73 #endif
  74
  75 #ifndef log2
  76 #define log2(n) cfs_ffz(~(n))
  77 #endif
  78
  79 static struct ll_sb_info *ll_init_sbi(void)
  80 {
  81         struct ll_sb_info *sbi = NULL;
  82         unsigned long pages;
  83         struct sysinfo si;
  84         class_uuid_t uuid;
  85         int i;
  86         ENTRY;
  87
  88         OBD_ALLOC(sbi, sizeof(*sbi));
  89         if (!sbi)
  90                 RETURN(NULL);
  91
  92         cfs_spin_lock_init(&sbi->ll_lock);
  93         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  94         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  95         cfs_spin_lock_init(&sbi->ll_process_lock);
  96         sbi->ll_rw_stats_on = 0;
  97
  98         si_meminfo(&si);
  99         pages = si.totalram - si.totalhigh;
 100         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
 101 #ifdef HAVE_BGL_SUPPORT
 102                 sbi->ll_async_page_max = pages / 4;
 103 #else
 104                 sbi->ll_async_page_max = pages / 2;
 105 #endif
 106         } else {
 107                 sbi->ll_async_page_max = (pages / 4) * 3;
 108         }
 109
 110         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 111                                            SBI_DEFAULT_READAHEAD_MAX);
 112         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 113         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 114                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 115         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 116         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 117
 118         ll_generate_random_uuid(uuid);
 119         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 120         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 121
 122         cfs_spin_lock(&ll_sb_lock);
 123         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 124         cfs_spin_unlock(&ll_sb_lock);
 125
 126 #ifdef ENABLE_CHECKSUM
 127         sbi->ll_flags |= LL_SBI_CHECKSUM;
 128 #endif
 129
 130 #ifdef HAVE_LRU_RESIZE_SUPPORT
 131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 132 #endif
 133
 134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 135                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 136                                    pp_r_hist.oh_lock);
 137                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 138                                    pp_w_hist.oh_lock);
 139         }
 140
 141         /* metadata statahead is enabled by default */
 142         sbi->ll_sa_max = LL_SA_RPC_DEF;
 143         atomic_set(&sbi->ll_sa_total, 0);
 144         atomic_set(&sbi->ll_sa_wrong, 0);
 145
 146         RETURN(sbi);
 147 }
 148
 149 void ll_free_sbi(struct super_block *sb)
 150 {
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         ENTRY;
 153
 154         if (sbi != NULL) {
 155                 cfs_spin_lock(&ll_sb_lock);
 156                 cfs_list_del(&sbi->ll_list);
 157                 cfs_spin_unlock(&ll_sb_lock);
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static struct dentry_operations ll_d_root_ops = {
 164         .d_compare = ll_dcompare,
 165         .d_revalidate = ll_revalidate_nd,
 166 };
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = 0;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_device *obd;
 174         struct obd_capa *oc = NULL;
 175         struct obd_statfs osfs;
 176         struct ptlrpc_request *request = NULL;
 177         struct obd_connect_data *data = NULL;
 178         struct obd_uuid *uuid;
 179         struct md_op_data *op_data;
 180         struct lustre_md lmd;
 181         obd_valid valid;
 182         int size, err, checksum;
 183         ENTRY;
 184
 185         obd = class_name2obd(md);
 186         if (!obd) {
 187                 CERROR("MD %s: not setup or attached\n", md);
 188                 RETURN(-EINVAL);
 189         }
 190
 191         OBD_ALLOC_PTR(data);
 192         if (data == NULL)
 193                 RETURN(-ENOMEM);
 194
 195         if (proc_lustre_fs_root) {
 196                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 197                                                   dt, md);
 198                 if (err < 0)
 199                         CERROR("could not register mount in /proc/fs/lustre\n");
 200         }
 201
 202         /* indicate the features supported by this client */
 203         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 204                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 205                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 206                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 207                                   OBD_CONNECT_CANCELSET| OBD_CONNECT_FID      |
 208                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 209                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 210                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH;
 211
 212         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 213                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 214
 215 #ifdef HAVE_LRU_RESIZE_SUPPORT
 216         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 217                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 218 #endif
 219 #ifdef CONFIG_FS_POSIX_ACL
 220         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 221 #endif
 222         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 223         data->ocd_version = LUSTRE_VERSION_CODE;
 224
 225         if (sb->s_flags & MS_RDONLY)
 226                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 227         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 228                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 229
 230 #ifdef HAVE_MS_FLOCK_LOCK
 231         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 232         sb->s_flags |= MS_FLOCK_LOCK;
 233 #endif
 234 #ifdef MS_HAS_NEW_AOPS
 235         sb->s_flags |= MS_HAS_NEW_AOPS;
 236 #endif
 237
 238         if (sbi->ll_flags & LL_SBI_FLOCK)
 239                 sbi->ll_fop = &ll_file_operations_flock;
 240         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 241                 sbi->ll_fop = &ll_file_operations;
 242         else
 243                 sbi->ll_fop = &ll_file_operations_noflock;
 244
 245         /* real client */
 246         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 247         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 248                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 249
 250         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 251
 252         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 253         if (err == -EBUSY) {
 254                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 255                                    "recovery, of which this client is not a "
 256                                    "part. Please wait for recovery to complete,"
 257                                    " abort, or time out.\n", md);
 258                 GOTO(out, err);
 259         } else if (err) {
 260                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 261                 GOTO(out, err);
 262         }
 263
 264         err = obd_fid_init(sbi->ll_md_exp);
 265         if (err) {
 266                 CERROR("Can't init metadata layer FID infrastructure, "
 267                        "rc %d\n", err);
 268                 GOTO(out_md, err);
 269         }
 270
 271         err = obd_statfs(obd, &osfs,
 272                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 273         if (err)
 274                 GOTO(out_md_fid, err);
 275
 276         size = sizeof(*data);
 277         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 278                            KEY_CONN_DATA,  &size, data, NULL);
 279         if (err) {
 280                 CERROR("Get connect data failed: %d \n", err);
 281                 GOTO(out_md, err);
 282         }
 283
 284         LASSERT(osfs.os_bsize);
 285         sb->s_blocksize = osfs.os_bsize;
 286         sb->s_blocksize_bits = log2(osfs.os_bsize);
 287         sb->s_magic = LL_SUPER_MAGIC;
 288
 289         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 290          *         retval = in_file->f_op->sendfile(...);
 291          *         if (*ppos > max)
 292          *                 retval = -EOVERFLOW;
 293          *
 294          * it will check if *ppos is greater than max. However, max equals to
 295          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 296          * has been defined as a signed long long integer in linux kernel. */
 297 #if BITS_PER_LONG == 64
 298         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 299 #else
 300         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 301 #endif
 302         sbi->ll_namelen = osfs.os_namelen;
 303         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 304
 305         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 306             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 307                 LCONSOLE_INFO("Disabling user_xattr feature because "
 308                               "it is not supported on the server\n");
 309                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 310         }
 311
 312         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 313 #ifdef MS_POSIXACL
 314                 sb->s_flags |= MS_POSIXACL;
 315 #endif
 316                 sbi->ll_flags |= LL_SBI_ACL;
 317         } else {
 318                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 319 #ifdef MS_POSIXACL
 320                 sb->s_flags &= ~MS_POSIXACL;
 321 #endif
 322                 sbi->ll_flags &= ~LL_SBI_ACL;
 323         }
 324
 325         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 326                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 327                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 328                         LCONSOLE_INFO("client is set as remote by default.\n");
 329                 }
 330         } else {
 331                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 332                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 333                         LCONSOLE_INFO("client claims to be remote, but server "
 334                                       "rejected, forced to be local.\n");
 335                 }
 336         }
 337
 338         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 339                 LCONSOLE_INFO("client enabled MDS capability!\n");
 340                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 341         }
 342
 343         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 344                 LCONSOLE_INFO("client enabled OSS capability!\n");
 345                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 346         }
 347
 348         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 349                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 350
 351         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 352                 sbi->ll_md_brw_size = data->ocd_brw_size;
 353         else
 354                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 355
 356         obd = class_name2obd(dt);
 357         if (!obd) {
 358                 CERROR("DT %s: not setup or attached\n", dt);
 359                 GOTO(out_md_fid, err = -ENODEV);
 360         }
 361
 362         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 363                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 364                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 365                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 366                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 367                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 368                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 369                                   OBD_CONNECT_MAXBYTES;
 370
 371         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 372                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 373
 374         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 375                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 376                  * disabled by default, because it can still be enabled on the
 377                  * fly via /proc. As a consequence, we still need to come to an
 378                  * agreement on the supported algorithms at connect time */
 379                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 380
 381                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 382                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 383                 else
 384                         /* send the list of supported checksum types */
 385                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 386         }
 387
 388 #ifdef HAVE_LRU_RESIZE_SUPPORT
 389         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 390 #endif
 391         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 392                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 393
 394         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 395                "ocd_grant: %d\n", data->ocd_connect_flags,
 396                data->ocd_version, data->ocd_grant);
 397
 398         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 399         obd->obd_upcall.onu_upcall = cl_ocd_update;
 400
 401         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 402
 403         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 404         if (err == -EBUSY) {
 405                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 406                                    "recovery, of which this client is not a "
 407                                    "part.  Please wait for recovery to "
 408                                    "complete, abort, or time out.\n", dt);
 409                 GOTO(out_md_fid, err);
 410         } else if (err) {
 411                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 412                 GOTO(out_md_fid, err);
 413         }
 414
 415         err = obd_fid_init(sbi->ll_dt_exp);
 416         if (err) {
 417                 CERROR("Can't init data layer FID infrastructure, "
 418                        "rc %d\n", err);
 419                 GOTO(out_dt, err);
 420         }
 421
 422         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 423         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 424         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 425         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 426         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 427
 428         fid_zero(&sbi->ll_root_fid);
 429         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 430         if (err) {
 431                 CERROR("cannot mds_connect: rc = %d\n", err);
 432                 GOTO(out_lock_cn_cb, err);
 433         }
 434         if (!fid_is_sane(&sbi->ll_root_fid)) {
 435                 CERROR("Invalid root fid during mount\n");
 436                 GOTO(out_lock_cn_cb, err = -EINVAL);
 437         }
 438         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 439
 440         sb->s_op = &lustre_super_operations;
 441 #if THREAD_SIZE >= 8192
 442         sb->s_export_op = &lustre_export_operations;
 443 #endif
 444
 445         /* make root inode
 446          * XXX: move this to after cbd setup? */
 447         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 448         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 449                 valid |= OBD_MD_FLRMTPERM;
 450         else if (sbi->ll_flags & LL_SBI_ACL)
 451                 valid |= OBD_MD_FLACL;
 452
 453         OBD_ALLOC_PTR(op_data);
 454         if (op_data == NULL)
 455                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 456
 457         op_data->op_fid1 = sbi->ll_root_fid;
 458         op_data->op_mode = 0;
 459         op_data->op_capa1 = oc;
 460         op_data->op_valid = valid;
 461
 462         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 463         if (oc)
 464                 capa_put(oc);
 465         OBD_FREE_PTR(op_data);
 466         if (err) {
 467                 CERROR("md_getattr failed for root: rc = %d\n", err);
 468                 GOTO(out_lock_cn_cb, err);
 469         }
 470         memset(&lmd, 0, sizeof(lmd));
 471         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 472                                sbi->ll_md_exp, &lmd);
 473         if (err) {
 474                 CERROR("failed to understand root inode md: rc = %d\n", err);
 475                 ptlrpc_req_finished (request);
 476                 GOTO(out_lock_cn_cb, err);
 477         }
 478
 479         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 480         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 481         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 482         ptlrpc_req_finished(request);
 483
 484         if (root == NULL || IS_ERR(root)) {
 485                 if (lmd.lsm)
 486                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 487 #ifdef CONFIG_FS_POSIX_ACL
 488                 if (lmd.posix_acl) {
 489                         posix_acl_release(lmd.posix_acl);
 490                         lmd.posix_acl = NULL;
 491                 }
 492 #endif
 493                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 494                 root = NULL;
 495                 CERROR("lustre_lite: bad iget4 for root\n");
 496                 GOTO(out_root, err);
 497         }
 498
 499         err = ll_close_thread_start(&sbi->ll_lcq);
 500         if (err) {
 501                 CERROR("cannot start close thread: rc %d\n", err);
 502                 GOTO(out_root, err);
 503         }
 504
 505 #ifdef CONFIG_FS_POSIX_ACL
 506         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 507                 rct_init(&sbi->ll_rct);
 508                 et_init(&sbi->ll_et);
 509         }
 510 #endif
 511
 512         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 513         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 514                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 515                                  NULL);
 516         cl_sb_init(sb);
 517
 518         sb->s_root = d_alloc_root(root);
 519         if (data != NULL)
 520                 OBD_FREE(data, sizeof(*data));
 521
 522         sb->s_root->d_op = &ll_d_root_ops;
 523
 524         sbi->ll_sdev_orig = sb->s_dev;
 525
 526         /* We set sb->s_dev equal on all lustre clients in order to support
 527          * NFS export clustering.  NFSD requires that the FSID be the same
 528          * on all clients. */
 529         /* s_dev is also used in lt_compare() to compare two fs, but that is
 530          * only a node-local comparison. */
 531         uuid = obd_get_uuid(sbi->ll_md_exp);
 532         if (uuid != NULL)
 533                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 534         sbi->ll_mnt = mnt;
 535
 536         RETURN(err);
 537 out_root:
 538         if (root)
 539                 iput(root);
 540 out_lock_cn_cb:
 541         obd_fid_fini(sbi->ll_dt_exp);
 542 out_dt:
 543         obd_disconnect(sbi->ll_dt_exp);
 544         sbi->ll_dt_exp = NULL;
 545 out_md_fid:
 546         obd_fid_fini(sbi->ll_md_exp);
 547 out_md:
 548         obd_disconnect(sbi->ll_md_exp);
 549         sbi->ll_md_exp = NULL;
 550 out:
 551         if (data != NULL)
 552                 OBD_FREE_PTR(data);
 553         lprocfs_unregister_mountpoint(sbi);
 554         return err;
 555 }
 556
 557 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 558 {
 559         int size, rc;
 560
 561         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 562         size = sizeof(int);
 563         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 564                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 565         if (rc)
 566                 CERROR("Get max mdsize error rc %d \n", rc);
 567
 568         RETURN(rc);
 569 }
 570
 571 void ll_dump_inode(struct inode *inode)
 572 {
 573         struct list_head *tmp;
 574         int dentry_count = 0;
 575
 576         LASSERT(inode != NULL);
 577
 578         list_for_each(tmp, &inode->i_dentry)
 579                 dentry_count++;
 580
 581         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 582                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 583                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 584 }
 585
 586 void lustre_dump_dentry(struct dentry *dentry, int recur)
 587 {
 588         struct list_head *tmp;
 589         int subdirs = 0;
 590
 591         LASSERT(dentry != NULL);
 592
 593         list_for_each(tmp, &dentry->d_subdirs)
 594                 subdirs++;
 595
 596         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 597                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 598                dentry->d_name.len, dentry->d_name.name,
 599                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 600                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 601                dentry->d_flags, dentry->d_fsdata, subdirs);
 602         if (dentry->d_inode != NULL)
 603                 ll_dump_inode(dentry->d_inode);
 604
 605         if (recur == 0)
 606                 return;
 607
 608        list_for_each(tmp, &dentry->d_subdirs) {
 609                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 610                 lustre_dump_dentry(d, recur - 1);
 611         }
 612 }
 613
 614 void client_common_put_super(struct super_block *sb)
 615 {
 616         struct ll_sb_info *sbi = ll_s2sbi(sb);
 617         ENTRY;
 618
 619 #ifdef CONFIG_FS_POSIX_ACL
 620         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 621                 et_fini(&sbi->ll_et);
 622                 rct_fini(&sbi->ll_rct);
 623         }
 624 #endif
 625
 626         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 627
 628         ll_close_thread_shutdown(sbi->ll_lcq);
 629
 630         cl_sb_fini(sb);
 631
 632         cfs_list_del(&sbi->ll_conn_chain);
 633
 634         obd_fid_fini(sbi->ll_dt_exp);
 635         obd_disconnect(sbi->ll_dt_exp);
 636         sbi->ll_dt_exp = NULL;
 637
 638         lprocfs_unregister_mountpoint(sbi);
 639
 640         obd_fid_fini(sbi->ll_md_exp);
 641         obd_disconnect(sbi->ll_md_exp);
 642         sbi->ll_md_exp = NULL;
 643
 644         EXIT;
 645 }
 646
 647 void ll_kill_super(struct super_block *sb)
 648 {
 649         struct ll_sb_info *sbi;
 650
 651         ENTRY;
 652
 653         /* not init sb ?*/
 654         if (!(sb->s_flags & MS_ACTIVE))
 655                 return;
 656
 657         sbi = ll_s2sbi(sb);
 658         /* we need restore s_dev from changed for clustred NFS before put_super
 659          * because new kernels have cached s_dev and change sb->s_dev in
 660          * put_super not affected real removing devices */
 661         if (sbi)
 662                 sb->s_dev = sbi->ll_sdev_orig;
 663         EXIT;
 664 }
 665
 666 char *ll_read_opt(const char *opt, char *data)
 667 {
 668         char *value;
 669         char *retval;
 670         ENTRY;
 671
 672         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 673         if (strncmp(opt, data, strlen(opt)))
 674                 RETURN(NULL);
 675         if ((value = strchr(data, '=')) == NULL)
 676                 RETURN(NULL);
 677
 678         value++;
 679         OBD_ALLOC(retval, strlen(value) + 1);
 680         if (!retval) {
 681                 CERROR("out of memory!\n");
 682                 RETURN(NULL);
 683         }
 684
 685         memcpy(retval, value, strlen(value)+1);
 686         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 687         RETURN(retval);
 688 }
 689
 690 static inline int ll_set_opt(const char *opt, char *data, int fl)
 691 {
 692         if (strncmp(opt, data, strlen(opt)) != 0)
 693                 return(0);
 694         else
 695                 return(fl);
 696 }
 697
 698 /* non-client-specific mount options are parsed in lmd_parse */
 699 static int ll_options(char *options, int *flags)
 700 {
 701         int tmp;
 702         char *s1 = options, *s2;
 703         ENTRY;
 704
 705         if (!options)
 706                 RETURN(0);
 707
 708         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 709
 710         while (*s1) {
 711                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 712                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 713                 if (tmp) {
 714                         *flags |= tmp;
 715                         goto next;
 716                 }
 717                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 718                 if (tmp) {
 719                         *flags |= tmp;
 720                         goto next;
 721                 }
 722                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 723                 if (tmp) {
 724                         *flags |= tmp;
 725                         goto next;
 726                 }
 727                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 728                 if (tmp) {
 729                         *flags &= ~tmp;
 730                         goto next;
 731                 }
 732                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 733                 if (tmp) {
 734                         *flags |= tmp;
 735                         goto next;
 736                 }
 737                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 738                 if (tmp) {
 739                         *flags &= ~tmp;
 740                         goto next;
 741                 }
 742 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 743                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 744                 if (tmp) {
 745                         /* Ignore deprecated mount option.  The client will
 746                          * always try to mount with ACL support, whether this
 747                          * is used depends on whether server supports it. */
 748                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 749                                                   "mount option 'acl'.\n");
 750                         goto next;
 751                 }
 752                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 753                 if (tmp) {
 754                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 755                                                   "mount option 'noacl'.\n");
 756                         goto next;
 757                 }
 758 #else
 759 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 760 #endif
 761                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 762                 if (tmp) {
 763                         *flags |= tmp;
 764                         goto next;
 765                 }
 766
 767                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 768                 if (tmp) {
 769                         *flags |= tmp;
 770                         goto next;
 771                 }
 772                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 773                 if (tmp) {
 774                         *flags &= ~tmp;
 775                         goto next;
 776                 }
 777                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 778                 if (tmp) {
 779                         *flags |= tmp;
 780                         goto next;
 781                 }
 782                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 783                 if (tmp) {
 784                         *flags &= ~tmp;
 785                         goto next;
 786                 }
 787                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 788                 if (tmp) {
 789                         *flags |= tmp;
 790                         goto next;
 791                 }
 792                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 793                 if (tmp) {
 794                         *flags &= ~tmp;
 795                         goto next;
 796                 }
 797                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 798                 if (tmp) {
 799                         *flags |= tmp;
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 803                 if (tmp) {
 804                         *flags |= tmp;
 805                         goto next;
 806                 }
 807
 808                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 809                                    s1);
 810                 RETURN(-EINVAL);
 811
 812 next:
 813                 /* Find next opt */
 814                 s2 = strchr(s1, ',');
 815                 if (s2 == NULL)
 816                         break;
 817                 s1 = s2 + 1;
 818         }
 819         RETURN(0);
 820 }
 821
 822 void ll_lli_init(struct ll_inode_info *lli)
 823 {
 824         lli->lli_inode_magic = LLI_INODE_MAGIC;
 825         cfs_sema_init(&lli->lli_size_sem, 1);
 826         cfs_sema_init(&lli->lli_write_sem, 1);
 827         cfs_init_rwsem(&lli->lli_trunc_sem);
 828         lli->lli_flags = 0;
 829         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 830         cfs_spin_lock_init(&lli->lli_lock);
 831         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 832         lli->lli_inode_magic = LLI_INODE_MAGIC;
 833         cfs_sema_init(&lli->lli_och_sem, 1);
 834         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 835         lli->lli_mds_exec_och = NULL;
 836         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 837         lli->lli_open_fd_exec_count = 0;
 838         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
 839         lli->lli_remote_perms = NULL;
 840         lli->lli_rmtperm_utime = 0;
 841         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 842         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 843         cfs_spin_lock_init(&lli->lli_sa_lock);
 844         cfs_sema_init(&lli->lli_readdir_sem, 1);
 845         fid_zero(&lli->lli_pfid);
 846 }
 847
 848 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 849 {
 850 #ifdef HAVE_BDI_REGISTER
 851         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 852
 853         bdi->name = "lustre";
 854         return bdi_register(bdi, NULL, "lustre-%d",
 855                             atomic_inc_return(&ll_bdi_num));
 856 #else
 857         return 0;
 858 #endif
 859 }
 860
 861 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 862 {
 863         struct lustre_profile *lprof;
 864         struct lustre_sb_info *lsi = s2lsi(sb);
 865         struct ll_sb_info *sbi;
 866         char  *dt = NULL, *md = NULL;
 867         char  *profilenm = get_profile_name(sb);
 868         struct config_llog_instance cfg = {0, };
 869         char   ll_instance[sizeof(sb) * 2 + 1];
 870         int    err;
 871         ENTRY;
 872
 873         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 874
 875         cfs_module_get();
 876
 877         /* client additional sb info */
 878         lsi->lsi_llsbi = sbi = ll_init_sbi();
 879         if (!sbi) {
 880                 cfs_module_put(THIS_MODULE);
 881                 RETURN(-ENOMEM);
 882         }
 883
 884         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 885         if (err)
 886                 GOTO(out_free, err);
 887
 888         err = ll_bdi_init(&lsi->lsi_bdi);
 889         if (err)
 890                 GOTO(out_free, err);
 891         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 892         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 893         err = ll_bdi_register(&lsi->lsi_bdi);
 894         if (err)
 895                 GOTO(out_free, err);
 896
 897 #ifdef HAVE_SB_BDI
 898         sb->s_bdi = &lsi->lsi_bdi;
 899 #endif
 900
 901         /* Generate a string unique to this super, in case some joker tries
 902            to mount the same fs at two mount points.
 903            Use the address of the super itself.*/
 904         sprintf(ll_instance, "%p", sb);
 905         cfg.cfg_instance = ll_instance;
 906         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 907
 908         /* set up client obds */
 909         err = lustre_process_log(sb, profilenm, &cfg);
 910         if (err < 0) {
 911                 CERROR("Unable to process log: %d\n", err);
 912                 GOTO(out_free, err);
 913         }
 914
 915         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 916         lprof = class_get_profile(profilenm);
 917         if (lprof == NULL) {
 918                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 919                                    " read from the MGS.  Does that filesystem "
 920                                    "exist?\n", profilenm);
 921                 GOTO(out_free, err = -EINVAL);
 922         }
 923         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 924                lprof->lp_md, lprof->lp_dt);
 925
 926         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 927                   strlen(ll_instance) + 2);
 928         if (!dt)
 929                 GOTO(out_free, err = -ENOMEM);
 930         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 931
 932         OBD_ALLOC(md, strlen(lprof->lp_md) +
 933                   strlen(ll_instance) + 2);
 934         if (!md)
 935                 GOTO(out_free, err = -ENOMEM);
 936         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 937
 938         /* connections, registrations, sb setup */
 939         err = client_common_fill_super(sb, md, dt, mnt);
 940
 941 out_free:
 942         if (md)
 943                 OBD_FREE(md, strlen(md) + 1);
 944         if (dt)
 945                 OBD_FREE(dt, strlen(dt) + 1);
 946         if (err)
 947                 ll_put_super(sb);
 948         else
 949                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 950
 951         RETURN(err);
 952 } /* ll_fill_super */
 953
 954
 955 void lu_context_keys_dump(void);
 956
 957 void ll_put_super(struct super_block *sb)
 958 {
 959         struct config_llog_instance cfg;
 960         char   ll_instance[sizeof(sb) * 2 + 1];
 961         struct obd_device *obd;
 962         struct lustre_sb_info *lsi = s2lsi(sb);
 963         struct ll_sb_info *sbi = ll_s2sbi(sb);
 964         char *profilenm = get_profile_name(sb);
 965         int force = 1, next;
 966         ENTRY;
 967
 968         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 969
 970         ll_print_capa_stat(sbi);
 971
 972         sprintf(ll_instance, "%p", sb);
 973         cfg.cfg_instance = ll_instance;
 974         lustre_end_log(sb, NULL, &cfg);
 975
 976         if (sbi->ll_md_exp) {
 977                 obd = class_exp2obd(sbi->ll_md_exp);
 978                 if (obd)
 979                         force = obd->obd_force;
 980         }
 981
 982         /* We need to set force before the lov_disconnect in
 983            lustre_common_put_super, since l_d cleans up osc's as well. */
 984         if (force) {
 985                 next = 0;
 986                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 987                                                      &next)) != NULL) {
 988                         obd->obd_force = force;
 989                 }
 990         }
 991
 992         if (sbi->ll_lcq) {
 993                 /* Only if client_common_fill_super succeeded */
 994                 client_common_put_super(sb);
 995         }
 996
 997         next = 0;
 998         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 999                 class_manual_cleanup(obd);
1000         }
1001
1002         if (profilenm)
1003                 class_del_profile(profilenm);
1004
1005         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1006                 ll_bdi_destroy(&lsi->lsi_bdi);
1007                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1008         }
1009
1010         ll_free_sbi(sb);
1011         lsi->lsi_llsbi = NULL;
1012
1013         lustre_common_put_super(sb);
1014
1015         cl_env_cache_purge(~0);
1016
1017         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1018
1019         cfs_module_put(THIS_MODULE);
1020
1021         EXIT;
1022 } /* client_put_super */
1023
1024 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1025 {
1026         struct inode *inode = NULL;
1027         /* NOTE: we depend on atomic igrab() -bzzz */
1028         lock_res_and_lock(lock);
1029         if (lock->l_ast_data) {
1030                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1031                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1032                         inode = igrab(lock->l_ast_data);
1033                 } else {
1034                         inode = lock->l_ast_data;
1035                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1036                                          D_WARNING, lock, "l_ast_data %p is "
1037                                          "bogus: magic %08x", lock->l_ast_data,
1038                                          lli->lli_inode_magic);
1039                         inode = NULL;
1040                 }
1041         }
1042         unlock_res_and_lock(lock);
1043         return inode;
1044 }
1045
1046 static int null_if_equal(struct ldlm_lock *lock, void *data)
1047 {
1048         if (data == lock->l_ast_data) {
1049                 lock->l_ast_data = NULL;
1050
1051                 if (lock->l_req_mode != lock->l_granted_mode)
1052                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1053         }
1054
1055         return LDLM_ITER_CONTINUE;
1056 }
1057
1058 void ll_clear_inode(struct inode *inode)
1059 {
1060         struct ll_inode_info *lli = ll_i2info(inode);
1061         struct ll_sb_info *sbi = ll_i2sbi(inode);
1062         ENTRY;
1063
1064         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1065                inode->i_generation, inode);
1066
1067         if (S_ISDIR(inode->i_mode)) {
1068                 /* these should have been cleared in ll_file_release */
1069                 LASSERT(lli->lli_sai == NULL);
1070                 LASSERT(lli->lli_opendir_key == NULL);
1071                 LASSERT(lli->lli_opendir_pid == 0);
1072         }
1073
1074         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1075         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1076                          null_if_equal, inode);
1077
1078         LASSERT(!lli->lli_open_fd_write_count);
1079         LASSERT(!lli->lli_open_fd_read_count);
1080         LASSERT(!lli->lli_open_fd_exec_count);
1081
1082         if (lli->lli_mds_write_och)
1083                 ll_md_real_close(inode, FMODE_WRITE);
1084         if (lli->lli_mds_exec_och)
1085                 ll_md_real_close(inode, FMODE_EXEC);
1086         if (lli->lli_mds_read_och)
1087                 ll_md_real_close(inode, FMODE_READ);
1088
1089         if (lli->lli_symlink_name) {
1090                 OBD_FREE(lli->lli_symlink_name,
1091                          strlen(lli->lli_symlink_name) + 1);
1092                 lli->lli_symlink_name = NULL;
1093         }
1094
1095         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1096                 LASSERT(lli->lli_posix_acl == NULL);
1097                 if (lli->lli_remote_perms) {
1098                         free_rmtperm_hash(lli->lli_remote_perms);
1099                         lli->lli_remote_perms = NULL;
1100                 }
1101         }
1102 #ifdef CONFIG_FS_POSIX_ACL
1103         else if (lli->lli_posix_acl) {
1104                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1105                 LASSERT(lli->lli_remote_perms == NULL);
1106                 posix_acl_release(lli->lli_posix_acl);
1107                 lli->lli_posix_acl = NULL;
1108         }
1109 #endif
1110         lli->lli_inode_magic = LLI_INODE_DEAD;
1111
1112         ll_clear_inode_capas(inode);
1113         /*
1114          * XXX This has to be done before lsm is freed below, because
1115          * cl_object still uses inode lsm.
1116          */
1117         cl_inode_fini(inode);
1118
1119         if (lli->lli_smd) {
1120                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1121                 lli->lli_smd = NULL;
1122         }
1123
1124
1125         EXIT;
1126 }
1127
1128 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1129                   struct md_open_data **mod)
1130 {
1131         struct lustre_md md;
1132         struct ll_sb_info *sbi = ll_i2sbi(inode);
1133         struct ptlrpc_request *request = NULL;
1134         int rc;
1135         ENTRY;
1136
1137         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1138                                      LUSTRE_OPC_ANY, NULL);
1139         if (IS_ERR(op_data))
1140                 RETURN(PTR_ERR(op_data));
1141
1142         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1143                         &request, mod);
1144         if (rc) {
1145                 ptlrpc_req_finished(request);
1146                 if (rc == -ENOENT) {
1147                         inode->i_nlink = 0;
1148                         /* Unlinked special device node? Or just a race?
1149                          * Pretend we done everything. */
1150                         if (!S_ISREG(inode->i_mode) &&
1151                             !S_ISDIR(inode->i_mode))
1152                                 rc = inode_setattr(inode, &op_data->op_attr);
1153                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1154                         CERROR("md_setattr fails: rc = %d\n", rc);
1155                 }
1156                 RETURN(rc);
1157         }
1158
1159         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1160                               sbi->ll_md_exp, &md);
1161         if (rc) {
1162                 ptlrpc_req_finished(request);
1163                 RETURN(rc);
1164         }
1165
1166         /* We call inode_setattr to adjust timestamps.
1167          * If there is at least some data in file, we cleared ATTR_SIZE
1168          * above to avoid invoking vmtruncate, otherwise it is important
1169          * to call vmtruncate in inode_setattr to update inode->i_size
1170          * (bug 6196) */
1171         rc = inode_setattr(inode, &op_data->op_attr);
1172
1173         /* Extract epoch data if obtained. */
1174         op_data->op_handle = md.body->handle;
1175         op_data->op_ioepoch = md.body->ioepoch;
1176
1177         ll_update_inode(inode, &md);
1178         ptlrpc_req_finished(request);
1179
1180         RETURN(rc);
1181 }
1182
1183 /* Close IO epoch and send Size-on-MDS attribute update. */
1184 static int ll_setattr_done_writing(struct inode *inode,
1185                                    struct md_op_data *op_data,
1186                                    struct md_open_data *mod)
1187 {
1188         struct ll_inode_info *lli = ll_i2info(inode);
1189         int rc = 0;
1190         ENTRY;
1191
1192         LASSERT(op_data != NULL);
1193         if (!S_ISREG(inode->i_mode))
1194                 RETURN(0);
1195
1196         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1197                op_data->op_ioepoch, PFID(&lli->lli_fid));
1198
1199         op_data->op_flags = MF_EPOCH_CLOSE;
1200         ll_done_writing_attr(inode, op_data);
1201         ll_pack_inode2opdata(inode, op_data, NULL);
1202
1203         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1204         if (rc == -EAGAIN) {
1205                 /* MDS has instructed us to obtain Size-on-MDS attribute
1206                  * from OSTs and send setattr to back to MDS. */
1207                 rc = ll_som_update(inode, op_data);
1208         } else if (rc) {
1209                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1210                        inode->i_ino, rc);
1211         }
1212         RETURN(rc);
1213 }
1214
1215 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1216 {
1217         struct obd_capa *capa;
1218         int rc;
1219
1220         if (attr->ia_valid & ATTR_SIZE)
1221                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1222         else
1223                 capa = ll_mdscapa_get(inode);
1224
1225         rc = cl_setattr_ost(inode, attr, capa);
1226
1227         if (attr->ia_valid & ATTR_SIZE)
1228                 ll_truncate_free_capa(capa);
1229         else
1230                 capa_put(capa);
1231
1232         return rc;
1233 }
1234
1235 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1236  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1237  * keep these values until such a time that objects are allocated for it.
1238  * We do the MDS operations first, as it is checking permissions for us.
1239  * We don't to the MDS RPC if there is nothing that we want to store there,
1240  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1241  * going to do an RPC anyways.
1242  *
1243  * If we are doing a truncate, we will send the mtime and ctime updates
1244  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1245  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1246  * at the same time.
1247  */
1248 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1249 {
1250         struct ll_inode_info *lli = ll_i2info(inode);
1251         struct lov_stripe_md *lsm = lli->lli_smd;
1252         struct md_op_data *op_data = NULL;
1253         struct md_open_data *mod = NULL;
1254         int ia_valid = attr->ia_valid;
1255         int rc = 0, rc1 = 0;
1256         ENTRY;
1257
1258         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1259                attr->ia_valid);
1260         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1261
1262         if (ia_valid & ATTR_SIZE) {
1263                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1264                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1265                                attr->ia_size, ll_file_maxbytes(inode));
1266                         RETURN(-EFBIG);
1267                 }
1268
1269                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1270         }
1271
1272         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1273         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1274                 if (cfs_curproc_fsuid() != inode->i_uid &&
1275                     !cfs_capable(CFS_CAP_FOWNER))
1276                         RETURN(-EPERM);
1277         }
1278
1279         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1280         if (attr->ia_valid & ATTR_CTIME) {
1281                 attr->ia_ctime = CFS_CURRENT_TIME;
1282                 attr->ia_valid |= ATTR_CTIME_SET;
1283         }
1284         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1285                 attr->ia_atime = CFS_CURRENT_TIME;
1286                 attr->ia_valid |= ATTR_ATIME_SET;
1287         }
1288         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1289                 attr->ia_mtime = CFS_CURRENT_TIME;
1290                 attr->ia_valid |= ATTR_MTIME_SET;
1291         }
1292
1293         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1294                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1295                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1296                        cfs_time_current_sec());
1297
1298         /* NB: ATTR_SIZE will only be set after this point if the size
1299          * resides on the MDS, ie, this file has no objects. */
1300         if (lsm)
1301                 attr->ia_valid &= ~ATTR_SIZE;
1302
1303         /* We always do an MDS RPC, even if we're only changing the size;
1304          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1305
1306         OBD_ALLOC_PTR(op_data);
1307         if (op_data == NULL)
1308                 RETURN(-ENOMEM);
1309
1310         UNLOCK_INODE_MUTEX(inode);
1311         if (ia_valid & ATTR_SIZE)
1312                 UP_WRITE_I_ALLOC_SEM(inode);
1313         cfs_down_write(&lli->lli_trunc_sem);
1314         LOCK_INODE_MUTEX(inode);
1315         if (ia_valid & ATTR_SIZE)
1316                 DOWN_WRITE_I_ALLOC_SEM(inode);
1317
1318         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1319
1320         /* Open epoch for truncate. */
1321         if (exp_connect_som(ll_i2mdexp(inode)) &&
1322             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1323                 op_data->op_flags = MF_EPOCH_OPEN;
1324
1325         rc = ll_md_setattr(inode, op_data, &mod);
1326         if (rc)
1327                 GOTO(out, rc);
1328
1329         ll_ioepoch_open(lli, op_data->op_ioepoch);
1330         if (!lsm || !S_ISREG(inode->i_mode)) {
1331                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1332                 GOTO(out, rc = 0);
1333         }
1334
1335         if (ia_valid & ATTR_SIZE)
1336                 attr->ia_valid |= ATTR_SIZE;
1337         if (ia_valid & (ATTR_SIZE |
1338                         ATTR_ATIME | ATTR_ATIME_SET |
1339                         ATTR_MTIME | ATTR_MTIME_SET))
1340                 /* on truncate and utimes send attributes to osts, setting
1341                  * mtime/atime to past will be performed under PW 0:EOF extent
1342                  * lock (new_size:EOF for truncate)
1343                  * it may seem excessive to send mtime/atime updates to osts
1344                  * when not setting times to past, but it is necessary due to
1345                  * possible time de-synchronization */
1346                 rc = ll_setattr_ost(inode, attr);
1347         EXIT;
1348 out:
1349         if (op_data) {
1350                 if (op_data->op_ioepoch)
1351                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1352                 ll_finish_md_op_data(op_data);
1353         }
1354         cfs_up_write(&lli->lli_trunc_sem);
1355         return rc ? rc : rc1;
1356 }
1357
1358 int ll_setattr(struct dentry *de, struct iattr *attr)
1359 {
1360         int mode = de->d_inode->i_mode;
1361
1362         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1363                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1364                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1365
1366         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1367                                (ATTR_SIZE|ATTR_MODE)) &&
1368             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1369              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1370               !(attr->ia_mode & S_ISGID))))
1371                 attr->ia_valid |= ATTR_FORCE;
1372
1373         if ((mode & S_ISUID) &&
1374             !(attr->ia_mode & S_ISUID) &&
1375             !(attr->ia_valid & ATTR_KILL_SUID))
1376                 attr->ia_valid |= ATTR_KILL_SUID;
1377
1378         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1379             !(attr->ia_mode & S_ISGID) &&
1380             !(attr->ia_valid & ATTR_KILL_SGID))
1381                 attr->ia_valid |= ATTR_KILL_SGID;
1382
1383         return ll_setattr_raw(de->d_inode, attr);
1384 }
1385
1386 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1387                        __u64 max_age, __u32 flags)
1388 {
1389         struct ll_sb_info *sbi = ll_s2sbi(sb);
1390         struct obd_statfs obd_osfs;
1391         int rc;
1392         ENTRY;
1393
1394         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1395         if (rc) {
1396                 CERROR("md_statfs fails: rc = %d\n", rc);
1397                 RETURN(rc);
1398         }
1399
1400         osfs->os_type = sb->s_magic;
1401
1402         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1403                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1404
1405         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1406                 flags |= OBD_STATFS_NODELAY;
1407
1408         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1409                               &obd_osfs, max_age, flags);
1410         if (rc) {
1411                 CERROR("obd_statfs fails: rc = %d\n", rc);
1412                 RETURN(rc);
1413         }
1414
1415         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1416                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1417                obd_osfs.os_files);
1418
1419         osfs->os_bsize = obd_osfs.os_bsize;
1420         osfs->os_blocks = obd_osfs.os_blocks;
1421         osfs->os_bfree = obd_osfs.os_bfree;
1422         osfs->os_bavail = obd_osfs.os_bavail;
1423
1424         /* If we don't have as many objects free on the OST as inodes
1425          * on the MDS, we reduce the total number of inodes to
1426          * compensate, so that the "inodes in use" number is correct.
1427          */
1428         if (obd_osfs.os_ffree < osfs->os_ffree) {
1429                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1430                         obd_osfs.os_ffree;
1431                 osfs->os_ffree = obd_osfs.os_ffree;
1432         }
1433
1434         RETURN(rc);
1435 }
1436 #ifndef HAVE_STATFS_DENTRY_PARAM
1437 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1438 {
1439 #else
1440 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1441 {
1442         struct super_block *sb = de->d_sb;
1443 #endif
1444         struct obd_statfs osfs;
1445         int rc;
1446
1447         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1448         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1449
1450         /* Some amount of caching on the client is allowed */
1451         rc = ll_statfs_internal(sb, &osfs,
1452                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1453                                 0);
1454         if (rc)
1455                 return rc;
1456
1457         statfs_unpack(sfs, &osfs);
1458
1459         /* We need to downshift for all 32-bit kernels, because we can't
1460          * tell if the kernel is being called via sys_statfs64() or not.
1461          * Stop before overflowing f_bsize - in which case it is better
1462          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1463         if (sizeof(long) < 8) {
1464                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1465                         sfs->f_bsize <<= 1;
1466
1467                         osfs.os_blocks >>= 1;
1468                         osfs.os_bfree >>= 1;
1469                         osfs.os_bavail >>= 1;
1470                 }
1471         }
1472
1473         sfs->f_blocks = osfs.os_blocks;
1474         sfs->f_bfree = osfs.os_bfree;
1475         sfs->f_bavail = osfs.os_bavail;
1476
1477         return 0;
1478 }
1479
1480 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1481 {
1482         struct ll_inode_info *lli;
1483         struct lov_stripe_md *lsm;
1484
1485         lli = ll_i2info(inode);
1486         LASSERT(lli->lli_size_sem_owner != current);
1487         cfs_down(&lli->lli_size_sem);
1488         LASSERT(lli->lli_size_sem_owner == NULL);
1489         lli->lli_size_sem_owner = current;
1490         lsm = lli->lli_smd;
1491         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1492                  lsm, lock_lsm);
1493         if (lock_lsm)
1494                 lov_stripe_lock(lsm);
1495 }
1496
1497 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1498 {
1499         struct ll_inode_info *lli;
1500         struct lov_stripe_md *lsm;
1501
1502         lli = ll_i2info(inode);
1503         lsm = lli->lli_smd;
1504         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1505                  lsm, unlock_lsm);
1506         if (unlock_lsm)
1507                 lov_stripe_unlock(lsm);
1508         LASSERT(lli->lli_size_sem_owner == current);
1509         lli->lli_size_sem_owner = NULL;
1510         cfs_up(&lli->lli_size_sem);
1511 }
1512
1513 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1514 {
1515         struct ll_inode_info *lli = ll_i2info(inode);
1516         struct mdt_body *body = md->body;
1517         struct lov_stripe_md *lsm = md->lsm;
1518         struct ll_sb_info *sbi = ll_i2sbi(inode);
1519
1520         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1521         if (lsm != NULL) {
1522                 cfs_down(&lli->lli_och_sem);
1523                 if (lli->lli_smd == NULL) {
1524                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1525                             lsm->lsm_magic != LOV_MAGIC_V3) {
1526                                 dump_lsm(D_ERROR, lsm);
1527                                 LBUG();
1528                         }
1529                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1530                                lsm, inode->i_ino, inode->i_generation, inode);
1531                         /* cl_inode_init must go before lli_smd or a race is
1532                          * possible where client thinks the file has stripes,
1533                          * but lov raid0 is not setup yet and parallel e.g.
1534                          * glimpse would try to use uninitialized lov */
1535                         cl_inode_init(inode, md);
1536                         cfs_spin_lock(&lli->lli_lock);
1537                         lli->lli_smd = lsm;
1538                         cfs_spin_unlock(&lli->lli_lock);
1539                         cfs_up(&lli->lli_och_sem);
1540                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1541                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1542                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1543                 } else {
1544                         cfs_up(&lli->lli_och_sem);
1545                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1546                                 lli->lli_smd->lsm_stripe_count ==
1547                                 lsm->lsm_stripe_count);
1548                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1549                                 CERROR("lsm mismatch for inode %ld\n",
1550                                        inode->i_ino);
1551                                 CERROR("lli_smd:\n");
1552                                 dump_lsm(D_ERROR, lli->lli_smd);
1553                                 CERROR("lsm:\n");
1554                                 dump_lsm(D_ERROR, lsm);
1555                                 LBUG();
1556                         }
1557                 }
1558                 if (lli->lli_smd != lsm)
1559                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1560         }
1561
1562         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1563                 if (body->valid & OBD_MD_FLRMTPERM)
1564                         ll_update_remote_perm(inode, md->remote_perm);
1565         }
1566 #ifdef CONFIG_FS_POSIX_ACL
1567         else if (body->valid & OBD_MD_FLACL) {
1568                 cfs_spin_lock(&lli->lli_lock);
1569                 if (lli->lli_posix_acl)
1570                         posix_acl_release(lli->lli_posix_acl);
1571                 lli->lli_posix_acl = md->posix_acl;
1572                 cfs_spin_unlock(&lli->lli_lock);
1573         }
1574 #endif
1575         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1576         inode->i_generation = cl_fid_build_gen(&body->fid1);
1577
1578         if (body->valid & OBD_MD_FLATIME) {
1579                 if (body->atime > LTIME_S(inode->i_atime))
1580                         LTIME_S(inode->i_atime) = body->atime;
1581                 lli->lli_lvb.lvb_atime = body->atime;
1582         }
1583         if (body->valid & OBD_MD_FLMTIME) {
1584                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1585                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1586                                "to "LPU64"\n", inode->i_ino,
1587                                LTIME_S(inode->i_mtime), body->mtime);
1588                         LTIME_S(inode->i_mtime) = body->mtime;
1589                 }
1590                 lli->lli_lvb.lvb_mtime = body->mtime;
1591         }
1592         if (body->valid & OBD_MD_FLCTIME) {
1593                 if (body->ctime > LTIME_S(inode->i_ctime))
1594                         LTIME_S(inode->i_ctime) = body->ctime;
1595                 lli->lli_lvb.lvb_ctime = body->ctime;
1596         }
1597         if (body->valid & OBD_MD_FLMODE)
1598                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1599         if (body->valid & OBD_MD_FLTYPE)
1600                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1601         LASSERT(inode->i_mode != 0);
1602         if (S_ISREG(inode->i_mode)) {
1603                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1604         } else {
1605                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1606         }
1607 #ifdef HAVE_INODE_BLKSIZE
1608         inode->i_blksize = 1<<inode->i_blkbits;
1609 #endif
1610         if (body->valid & OBD_MD_FLUID)
1611                 inode->i_uid = body->uid;
1612         if (body->valid & OBD_MD_FLGID)
1613                 inode->i_gid = body->gid;
1614         if (body->valid & OBD_MD_FLFLAGS)
1615                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1616         if (body->valid & OBD_MD_FLNLINK)
1617                 inode->i_nlink = body->nlink;
1618         if (body->valid & OBD_MD_FLRDEV)
1619                 inode->i_rdev = old_decode_dev(body->rdev);
1620
1621         if (body->valid & OBD_MD_FLID) {
1622                 /* FID shouldn't be changed! */
1623                 if (fid_is_sane(&lli->lli_fid)) {
1624                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1625                                  "Trying to change FID "DFID
1626                                  " to the "DFID", inode %lu/%u(%p)\n",
1627                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1628                                  inode->i_ino, inode->i_generation, inode);
1629                 } else
1630                         lli->lli_fid = body->fid1;
1631         }
1632
1633         LASSERT(fid_seq(&lli->lli_fid) != 0);
1634
1635         if (body->valid & OBD_MD_FLSIZE) {
1636                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1637                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1638                         struct lustre_handle lockh;
1639                         ldlm_mode_t mode;
1640
1641                         /* As it is possible a blocking ast has been processed
1642                          * by this time, we need to check there is an UPDATE
1643                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1644                          * it. */
1645                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1646                                                &lockh);
1647                         if (mode) {
1648                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1649                                                       LLIF_EPOCH_PENDING |
1650                                                       LLIF_SOM_DIRTY)) {
1651                                         CERROR("ino %lu flags %lu still has "
1652                                                "size authority! do not trust "
1653                                                "the size got from MDS\n",
1654                                                inode->i_ino, lli->lli_flags);
1655                                 } else {
1656                                         /* Use old size assignment to avoid
1657                                          * deadlock bz14138 & bz14326 */
1658                                         inode->i_size = body->size;
1659                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1660                                 }
1661                                 ldlm_lock_decref(&lockh, mode);
1662                         }
1663                 } else {
1664                         /* Use old size assignment to avoid
1665                          * deadlock bz14138 & bz14326 */
1666                         inode->i_size = body->size;
1667
1668                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1669                                inode->i_ino, (unsigned long long)body->size);
1670                 }
1671
1672                 if (body->valid & OBD_MD_FLBLOCKS)
1673                         inode->i_blocks = body->blocks;
1674         }
1675
1676         if (body->valid & OBD_MD_FLMDSCAPA) {
1677                 LASSERT(md->mds_capa);
1678                 ll_add_capa(inode, md->mds_capa);
1679         }
1680         if (body->valid & OBD_MD_FLOSSCAPA) {
1681                 LASSERT(md->oss_capa);
1682                 ll_add_capa(inode, md->oss_capa);
1683         }
1684 }
1685
1686 void ll_read_inode2(struct inode *inode, void *opaque)
1687 {
1688         struct lustre_md *md = opaque;
1689         struct ll_inode_info *lli = ll_i2info(inode);
1690         ENTRY;
1691
1692         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1693                inode->i_ino, inode->i_generation, inode);
1694
1695         ll_lli_init(lli);
1696
1697         LASSERT(!lli->lli_smd);
1698
1699         /* Core attributes from the MDS first.  This is a new inode, and
1700          * the VFS doesn't zero times in the core inode so we have to do
1701          * it ourselves.  They will be overwritten by either MDS or OST
1702          * attributes - we just need to make sure they aren't newer. */
1703         LTIME_S(inode->i_mtime) = 0;
1704         LTIME_S(inode->i_atime) = 0;
1705         LTIME_S(inode->i_ctime) = 0;
1706         inode->i_rdev = 0;
1707         ll_update_inode(inode, md);
1708
1709         /* OIDEBUG(inode); */
1710
1711         /* initializing backing dev info. */
1712         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1713
1714
1715         if (S_ISREG(inode->i_mode)) {
1716                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1717                 inode->i_op = &ll_file_inode_operations;
1718                 inode->i_fop = sbi->ll_fop;
1719                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1720                 EXIT;
1721         } else if (S_ISDIR(inode->i_mode)) {
1722                 inode->i_op = &ll_dir_inode_operations;
1723                 inode->i_fop = &ll_dir_operations;
1724                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1725                 EXIT;
1726         } else if (S_ISLNK(inode->i_mode)) {
1727                 inode->i_op = &ll_fast_symlink_inode_operations;
1728                 EXIT;
1729         } else {
1730                 inode->i_op = &ll_special_inode_operations;
1731
1732                 init_special_inode(inode, inode->i_mode,
1733                                    kdev_t_to_nr(inode->i_rdev));
1734
1735                 EXIT;
1736         }
1737 }
1738
1739 void ll_delete_inode(struct inode *inode)
1740 {
1741         struct ll_sb_info *sbi = ll_i2sbi(inode);
1742         int rc;
1743         ENTRY;
1744
1745         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1746         if (rc)
1747                 CERROR("fid_delete() failed, rc %d\n", rc);
1748
1749         truncate_inode_pages(&inode->i_data, 0);
1750
1751         /* Workaround for LU-118 */
1752         if (inode->i_data.nrpages) {
1753                 TREE_READ_LOCK_IRQ(&inode->i_data);
1754                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1755                 LASSERTF(inode->i_data.nrpages == 0,
1756                          "inode=%lu/%u(%p) nrpages=%lu, see "
1757                          "http://jira.whamcloud.com/browse/LU-118\n",
1758                          inode->i_ino, inode->i_generation, inode,
1759                          inode->i_data.nrpages);
1760         }
1761         /* Workaround end */
1762
1763         clear_inode(inode);
1764
1765         EXIT;
1766 }
1767
1768 int ll_iocontrol(struct inode *inode, struct file *file,
1769                  unsigned int cmd, unsigned long arg)
1770 {
1771         struct ll_sb_info *sbi = ll_i2sbi(inode);
1772         struct ptlrpc_request *req = NULL;
1773         int rc, flags = 0;
1774         ENTRY;
1775
1776         switch(cmd) {
1777         case FSFILT_IOC_GETFLAGS: {
1778                 struct mdt_body *body;
1779                 struct md_op_data *op_data;
1780
1781                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1782                                              0, 0, LUSTRE_OPC_ANY,
1783                                              NULL);
1784                 if (op_data == NULL)
1785                         RETURN(-ENOMEM);
1786
1787                 op_data->op_valid = OBD_MD_FLFLAGS;
1788                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1789                 ll_finish_md_op_data(op_data);
1790                 if (rc) {
1791                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1792                         RETURN(-abs(rc));
1793                 }
1794
1795                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1796
1797                 flags = body->flags;
1798
1799                 ptlrpc_req_finished(req);
1800
1801                 RETURN(put_user(flags, (int *)arg));
1802         }
1803         case FSFILT_IOC_SETFLAGS: {
1804                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1805                 struct obd_info oinfo = { { { 0 } } };
1806                 struct md_op_data *op_data;
1807
1808                 if (get_user(flags, (int *)arg))
1809                         RETURN(-EFAULT);
1810
1811                 oinfo.oi_md = lsm;
1812                 OBDO_ALLOC(oinfo.oi_oa);
1813                 if (!oinfo.oi_oa)
1814                         RETURN(-ENOMEM);
1815
1816                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1817                                              LUSTRE_OPC_ANY, NULL);
1818                 if (IS_ERR(op_data))
1819                         RETURN(PTR_ERR(op_data));
1820
1821                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1822                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1823                 rc = md_setattr(sbi->ll_md_exp, op_data,
1824                                 NULL, 0, NULL, 0, &req, NULL);
1825                 ll_finish_md_op_data(op_data);
1826                 ptlrpc_req_finished(req);
1827                 if (rc) {
1828                         OBDO_FREE(oinfo.oi_oa);
1829                         RETURN(rc);
1830                 }
1831
1832                 if (lsm == NULL) {
1833                         OBDO_FREE(oinfo.oi_oa);
1834                         GOTO(update_cache, rc);
1835                 }
1836
1837                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1838                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1839                 oinfo.oi_oa->o_flags = flags;
1840                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1841                                        OBD_MD_FLGROUP;
1842                 oinfo.oi_capa = ll_mdscapa_get(inode);
1843                 obdo_from_inode(oinfo.oi_oa, inode,
1844                                 &ll_i2info(inode)->lli_fid, 0);
1845                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1846                 capa_put(oinfo.oi_capa);
1847                 OBDO_FREE(oinfo.oi_oa);
1848                 if (rc) {
1849                         if (rc != -EPERM && rc != -EACCES)
1850                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1851                         RETURN(rc);
1852                 }
1853
1854                 EXIT;
1855 update_cache:
1856                 inode->i_flags = ll_ext_to_inode_flags(flags);
1857                 return 0;
1858         }
1859         default:
1860                 RETURN(-ENOSYS);
1861         }
1862
1863         RETURN(0);
1864 }
1865
1866 int ll_flush_ctx(struct inode *inode)
1867 {
1868         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1869
1870         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1871
1872         obd_set_info_async(sbi->ll_md_exp,
1873                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1874                            0, NULL, NULL);
1875         obd_set_info_async(sbi->ll_dt_exp,
1876                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1877                            0, NULL, NULL);
1878         return 0;
1879 }
1880
1881 /* umount -f client means force down, don't save state */
1882 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1883 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1884 {
1885         struct super_block *sb = vfsmnt->mnt_sb;
1886 #else
1887 void ll_umount_begin(struct super_block *sb)
1888 {
1889 #endif
1890         struct lustre_sb_info *lsi = s2lsi(sb);
1891         struct ll_sb_info *sbi = ll_s2sbi(sb);
1892         struct obd_device *obd;
1893         struct obd_ioctl_data *ioc_data;
1894         ENTRY;
1895
1896 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1897         if (!(flags & MNT_FORCE)) {
1898                 EXIT;
1899                 return;
1900         }
1901 #endif
1902
1903         /* Tell the MGC we got umount -f */
1904         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1905
1906         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1907                sb->s_count, atomic_read(&sb->s_active));
1908
1909         obd = class_exp2obd(sbi->ll_md_exp);
1910         if (obd == NULL) {
1911                 CERROR("Invalid MDC connection handle "LPX64"\n",
1912                        sbi->ll_md_exp->exp_handle.h_cookie);
1913                 EXIT;
1914                 return;
1915         }
1916         obd->obd_force = 1;
1917
1918         obd = class_exp2obd(sbi->ll_dt_exp);
1919         if (obd == NULL) {
1920                 CERROR("Invalid LOV connection handle "LPX64"\n",
1921                        sbi->ll_dt_exp->exp_handle.h_cookie);
1922                 EXIT;
1923                 return;
1924         }
1925         obd->obd_force = 1;
1926
1927         OBD_ALLOC_PTR(ioc_data);
1928         if (ioc_data) {
1929                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1930                               sizeof ioc_data, ioc_data, NULL);
1931
1932                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1933                               sizeof ioc_data, ioc_data, NULL);
1934
1935                 OBD_FREE_PTR(ioc_data);
1936         }
1937
1938
1939         /* Really, we'd like to wait until there are no requests outstanding,
1940          * and then continue.  For now, we just invalidate the requests,
1941          * schedule() and sleep one second if needed, and hope.
1942          */
1943         cfs_schedule();
1944 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1945         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1946                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1947                                                    cfs_time_seconds(1));
1948                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1949                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1950                                       "may try to umount it a bit later\n",
1951                                       atomic_read(&vfsmnt->mnt_count));
1952         }
1953 #endif
1954
1955         EXIT;
1956 }
1957
1958 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1959 {
1960         struct ll_sb_info *sbi = ll_s2sbi(sb);
1961         int err;
1962         __u32 read_only;
1963
1964         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1965                 read_only = *flags & MS_RDONLY;
1966                 err = obd_set_info_async(sbi->ll_md_exp,
1967                                          sizeof(KEY_READ_ONLY),
1968                                          KEY_READ_ONLY, sizeof(read_only),
1969                                          &read_only, NULL);
1970                 if (err) {
1971                         CERROR("Failed to change the read-only flag during "
1972                                "remount: %d\n", err);
1973                         return err;
1974                 }
1975
1976                 if (read_only)
1977                         sb->s_flags |= MS_RDONLY;
1978                 else
1979                         sb->s_flags &= ~MS_RDONLY;
1980         }
1981         return 0;
1982 }
1983
1984 int ll_prep_inode(struct inode **inode,
1985                   struct ptlrpc_request *req,
1986                   struct super_block *sb)
1987 {
1988         struct ll_sb_info *sbi = NULL;
1989         struct lustre_md md;
1990         int rc;
1991         ENTRY;
1992
1993         LASSERT(*inode || sb);
1994         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1995         memset(&md, 0, sizeof(struct lustre_md));
1996
1997         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1998                               sbi->ll_md_exp, &md);
1999         if (rc)
2000                 RETURN(rc);
2001
2002         if (*inode) {
2003                 ll_update_inode(*inode, &md);
2004         } else {
2005                 LASSERT(sb != NULL);
2006
2007                 /*
2008                  * At this point server returns to client's same fid as client
2009                  * generated for creating. So using ->fid1 is okay here.
2010                  */
2011                 LASSERT(fid_is_sane(&md.body->fid1));
2012
2013                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2014                 if (*inode == NULL || IS_ERR(*inode)) {
2015                         if (md.lsm)
2016                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2017 #ifdef CONFIG_FS_POSIX_ACL
2018                         if (md.posix_acl) {
2019                                 posix_acl_release(md.posix_acl);
2020                                 md.posix_acl = NULL;
2021                         }
2022 #endif
2023                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2024                         *inode = NULL;
2025                         CERROR("new_inode -fatal: rc %d\n", rc);
2026                         GOTO(out, rc);
2027                 }
2028         }
2029
2030 out:
2031         md_free_lustre_md(sbi->ll_md_exp, &md);
2032         RETURN(rc);
2033 }
2034
2035 int ll_obd_statfs(struct inode *inode, void *arg)
2036 {
2037         struct ll_sb_info *sbi = NULL;
2038         struct obd_export *exp;
2039         char *buf = NULL;
2040         struct obd_ioctl_data *data = NULL;
2041         __u32 type;
2042         int len = 0, rc;
2043
2044         if (!inode || !(sbi = ll_i2sbi(inode)))
2045                 GOTO(out_statfs, rc = -EINVAL);
2046
2047         rc = obd_ioctl_getdata(&buf, &len, arg);
2048         if (rc)
2049                 GOTO(out_statfs, rc);
2050
2051         data = (void*)buf;
2052         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2053             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2054                 GOTO(out_statfs, rc = -EINVAL);
2055
2056         if (data->ioc_inllen1 != sizeof(__u32) ||
2057             data->ioc_inllen2 != sizeof(__u32) ||
2058             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2059             data->ioc_plen2 != sizeof(struct obd_uuid))
2060                 GOTO(out_statfs, rc = -EINVAL);
2061
2062         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2063         if (type == LL_STATFS_MDC)
2064                 exp = sbi->ll_md_exp;
2065         else if (type == LL_STATFS_LOV)
2066                 exp = sbi->ll_dt_exp;
2067         else
2068                 GOTO(out_statfs, rc = -ENODEV);
2069
2070         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2071         if (rc)
2072                 GOTO(out_statfs, rc);
2073 out_statfs:
2074         if (buf)
2075                 obd_ioctl_freedata(buf, len);
2076         return rc;
2077 }
2078
2079 int ll_process_config(struct lustre_cfg *lcfg)
2080 {
2081         char *ptr;
2082         void *sb;
2083         struct lprocfs_static_vars lvars;
2084         unsigned long x;
2085         int rc = 0;
2086
2087         lprocfs_llite_init_vars(&lvars);
2088
2089         /* The instance name contains the sb: lustre-client-aacfe000 */
2090         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2091         if (!ptr || !*(++ptr))
2092                 return -EINVAL;
2093         if (sscanf(ptr, "%lx", &x) != 1)
2094                 return -EINVAL;
2095         sb = (void *)x;
2096         /* This better be a real Lustre superblock! */
2097         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2098
2099         /* Note we have not called client_common_fill_super yet, so
2100            proc fns must be able to handle that! */
2101         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2102                                       lcfg, sb);
2103         if (rc > 0)
2104                 rc = 0;
2105         return(rc);
2106 }
2107
2108 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2109 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2110                                        struct inode *i1, struct inode *i2,
2111                                        const char *name, int namelen,
2112                                        int mode, __u32 opc, void *data)
2113 {
2114         LASSERT(i1 != NULL);
2115
2116         if (namelen > ll_i2sbi(i1)->ll_namelen)
2117                 return ERR_PTR(-ENAMETOOLONG);
2118
2119         if (op_data == NULL)
2120                 OBD_ALLOC_PTR(op_data);
2121
2122         if (op_data == NULL)
2123                 return ERR_PTR(-ENOMEM);
2124
2125         ll_i2gids(op_data->op_suppgids, i1, i2);
2126         op_data->op_fid1 = *ll_inode2fid(i1);
2127         op_data->op_capa1 = ll_mdscapa_get(i1);
2128
2129         if (i2) {
2130                 op_data->op_fid2 = *ll_inode2fid(i2);
2131                 op_data->op_capa2 = ll_mdscapa_get(i2);
2132         } else {
2133                 fid_zero(&op_data->op_fid2);
2134                 op_data->op_capa2 = NULL;
2135         }
2136
2137         op_data->op_name = name;
2138         op_data->op_namelen = namelen;
2139         op_data->op_mode = mode;
2140         op_data->op_mod_time = cfs_time_current_sec();
2141         op_data->op_fsuid = cfs_curproc_fsuid();
2142         op_data->op_fsgid = cfs_curproc_fsgid();
2143         op_data->op_cap = cfs_curproc_cap_pack();
2144         op_data->op_bias = MDS_CHECK_SPLIT;
2145         op_data->op_opc = opc;
2146         op_data->op_mds = 0;
2147         op_data->op_data = data;
2148
2149         /* If the file is being opened after mknod() (normally due to NFS)
2150          * try to use the default stripe data from parent directory for
2151          * allocating OST objects.  Try to pass the parent FID to MDS. */
2152         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2153             ll_i2info(i2)->lli_smd == NULL) {
2154                 struct ll_inode_info *lli = ll_i2info(i2);
2155
2156                 cfs_spin_lock(&lli->lli_lock);
2157                 if (likely(lli->lli_smd == NULL &&
2158                            !fid_is_zero(&lli->lli_pfid)))
2159                         op_data->op_fid1 = lli->lli_pfid;
2160                 cfs_spin_unlock(&lli->lli_lock);
2161                 /** We ignore parent's capability temporary. */
2162         }
2163
2164         return op_data;
2165 }
2166
2167 void ll_finish_md_op_data(struct md_op_data *op_data)
2168 {
2169         capa_put(op_data->op_capa1);
2170         capa_put(op_data->op_capa2);
2171         OBD_FREE_PTR(op_data);
2172 }
2173
2174 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2175 {
2176         struct ll_sb_info *sbi;
2177
2178         LASSERT((seq != NULL) && (vfs != NULL));
2179         sbi = ll_s2sbi(vfs->mnt_sb);
2180
2181         if (sbi->ll_flags & LL_SBI_NOLCK)
2182                 seq_puts(seq, ",nolock");
2183
2184         if (sbi->ll_flags & LL_SBI_FLOCK)
2185                 seq_puts(seq, ",flock");
2186
2187         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2188                 seq_puts(seq, ",localflock");
2189
2190         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2191                 seq_puts(seq, ",user_xattr");
2192
2193         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2194                 seq_puts(seq, ",lazystatfs");
2195
2196         RETURN(0);
2197 }